Title: Enabling Personalized Prosthetic Control Using Self-learning and Bayesian Optimization
Date: Friday, January 24th, 2025
Time: 2:00 PM EST
Location: GTMI Auditorium
Zoom link: https://gatech.zoom.us/j/92464959824
Committee Members:
Dr. Aaron J. Young (Advisor), School of Mechanical Engineering, Georgia Institute of Technology
Dr. Nick Fey, Walker Department of Mechanical Engineering, University of Texas at Austin
Dr. Matthew Gombolay, School of Interactive Computing, Georgia Institute of Technology
Dr. Ye Zhao, School of Mechanical Engineering, Georgia Institute of Technology
Dr. Young-Hui Chang, School of Biological Sciences, Georgia Institute of Technology
Abstract:
The field of lower-limb prosthetics aims to restore mobility and enhance the quality of life for individuals with transtibial and transfemoral amputations. While powered prostheses have shown significant potential in replicating able-bodied biomechanics, their adoption remains limited by the critical need to provide task-appropriate and effective assistance. Current state-of-the-art systems rely heavily on offline, user-specific tuning and data collection, making them impractical for widespread use. This dissertation addresses these limitations through three key objectives: 1) developing real-time continual learning algorithms that enable machine learning models for walking speed estimation to self-improve in user-independent scenarios without requiring offline user-specific data, 2) leveraging transfer learning to pre-train models on able-bodied datasets and fine-tune them for transfemoral prosthesis users to compensate for the lack of prosthesis-specific data, and 3) applying Bayesian optimization to minimize gait asymmetry by identifying optimal knee and ankle control parameters across varying walking speeds. The results demonstrate that real-time continual learning significantly improves prosthetic performance within minutes of walking, transfer learning effectively bridges the gap between able-bodied and prosthesis-specific datasets, and Bayesian optimization ensures personalized assistance tailored to individual user needs. Together, these innovations advance powered prostheses toward safe and personalized functionality that works out of the box, enabling broader accessibility and improved mobility for diverse users and environments.